Facilitating message composition based on absent context

Abstract

Methods, computer systems, computer-storage media, and graphical user interfaces are provided for facilitating message composition, according to embodiments of the present invention. In one embodiment, message data associated with a message being composed is obtained. The message data is analyzed to determine a message type indicating a type of message and a message context representation representing a context provided within the message being composed. Context representations representing expected contexts associated with the message type of the message are identified. Thereafter, an absent context missing in the message being composed is determined based on a comparison of the message context representation with the set of context representations. A recommendation related to the absent context can be provided, for example, for display via a user interface.

What is claimed is: 
     
       1. A computing system comprising:
 a processor; and 
 computer storage memory having computer-executable instructions stored thereon which, when executed by the processor, configure the computing system to:
 obtain message data associated with a message being composed in a user interface displayed in a display; 
 based on the message data:
 cause a classification model to determine, based on the message data and a set of message types, a message type of the set of message types for the message being composed that designates the message as the message type, wherein the classification model is a multi-class classification model and the set of message types comprises: a report message type, an incident report message type, an information message type, a query-based message type, an insightful message type, and a commitment message type, wherein the set of message types are determined based on an output of the classification model; and 
 generate a message context representation indicating a context provided within the message being composed, the message context representation includes a first vector, generated via a deep neural network, representing the context provided within the message being composed and comprises an n-gram format that is a contiguous sequence of n items of text; 
 
 based on the message type of the message being composed, obtaining, from a context database, an expected context representation indicating an expected context associated with the message type of the message, the expected context comprising text that is expected to occur in messages of the message type of the message, where the expected context representation includes a second vector representing the expected context associated with the message type of the message and the expected context representation is identified via the context database, the context database including a record having an indication of the message type and a context representation corresponding to the indicated message type and is identified as the context representation corresponding to the indicated message type of the record; 
 determine an absent context missing in the message being composed based on a comparison of values included in the first vector and the second vector, the absent context is determined based on a similarity score of a first value from the first vector and a second value from the second vector being below a threshold; and 
 cause a recommendation related to the absent context missing in the message being composed to be displayed via the user interface. 
 
 
     
     
       2. The computing system of  claim 1 , wherein the message comprises an email and wherein the message data comprises message content, an indication of a message recipient, and a message subject line. 
     
     
       3. The computing system of  claim 1 , wherein a data format of the expected context representation data comprises an embedding format. 
     
     
       4. The computing system of  claim 1 , further comprising:
 determining, based on the message type, a context type of the message being composed, the context type comprising one of a knowledge context type, a task context type, an impact context type, and a question context type; and 
 wherein the message context representation is further determined based on the context type. 
 
     
     
       5. The computing system of  claim 1 , wherein the expected context representation is identified as the context representation corresponding to the indicated message type of the record. 
     
     
       6. The computing system of  claim 1 , wherein determining the absent context comprises determining similarity of the message context representation and the expected context representation. 
     
     
       7. The computing system of  claim 1 , wherein the recommendation provides an indication of the absent context missing in the message being composed. 
     
     
       8. The computing system of  claim 1 , wherein the message data comprises an indication of a message recipient, and wherein the message type is determined based on the indication of the message recipient. 
     
     
       9. A computer-implemented method for facilitating message composition, the method comprising:
 determining, via a multi-class classification model, based on message data associated with a message being composed in a user interface displayed in a display, a message type of a set of message types associated with the message being composed, wherein the set of message types comprises: a report message type, an incident report message type, an information message type, a query-based message type, an insightful message type, and a commitment message type, wherein the set of message types that are input into a context database are determined based on an output of the classification model; 
 based on the message type, identifying, via the context database, an expected context representation that is a first vector, generated via a deep neural network, indicating an expected context corresponding with the message type associated with the message being composed based on a first set of values included in the first vector comprising an n-gram format that represents a contiguous sequence of n items of text, wherein the expected context representation is identified via the context database, the context database including a record including the message type and a message context representation corresponding to the message type suitable for indicating the expected context representation; 
 determining that the expected context is absent from the message being composed based on a similarity analysis of the expected context representation with the message context representation that is a second vector indicating context within the message being composed, where the similarity analysis includes at least a comparison of a second set of values included in the second vector to the first set of values included in the first vector; and 
 causing a recommendation indicating that the expected context is absent from the message being composed to be displayed via the user interface. 
 
     
     
       10. The method of  claim 9 , wherein the similarity analysis comprises determining a similarity score indicating similarity between the expected context representation and the message context representation. 
     
     
       11. The method of  claim 9 , wherein the recommendation prompts a user to include the expected context in the message being composed. 
     
     
       12. The method of  claim 9 , wherein the set of message types are input into the context database via a graphical user interface. 
     
     
       13. The method of  claim 9 , wherein the expected context corresponding with the message type comprises text that is expected to occur in messages of the message type. 
     
     
       14. One or more computer storage media having computer-executable instructions embodied thereon that, when executed by one or more processors, cause the one or more processors to perform a method for facilitating error detection, the method comprising:
 generating a context database that includes a set of message types and a set of context representations, context representation of the set of context representation comprising a data format include a vector representation, generated via a deep neural network, indicating an expected context for a message type and comprising an n-gram format representing contiguous sequence text, the expected context comprising text that is expected to occur in messages of the message type; 
 causing a classification model to generate an indication of a first message type for a message being composed in a user interface displayed in a display based at least in part on message data associated with the message being composed, wherein the classification model is a multi-class classification model and the set of message types comprises: a report message type, an incident report message type, an information message type, a query-based message type, an insightful message type, and a commitment message type, wherein the set of message types are automatically determined via machine learning; 
 identifying, via the context database, a first context representation indicating a first expected context corresponding with the first message type indicated for the message being composed, where the first context representation includes a first vector of a first set of values, wherein the first expected context representation is identified based on a record included in the context database indicating the first message type corresponds to the first expected context; 
 determining that the first expected context is missing from the message being composed based on a comparison of the first context representation with a message context representation that includes a second vector indicating a context within the message being composed by at least comparing a second set of values of the second vector to the first set of values of the first vector; and 
 causing a recommendation related to the first expected context to be displayed via the user interface. 
 
     
     
       15. The media of  claim 14 , wherein the first message type for the message being composed is determined based on a recipient of the message being composed, and wherein at least one message types of the set of message types include in the context database corresponds with the recipient. 
     
     
       16. The media of  claim 14 , wherein at least a portion of the set of message types or the context representations are pre-determined and input into the context database via a graphical user interface. 
     
     
       17. The media of  claim 14 , wherein the context representations are automatically determined via machine learning. 
     
     
       18. The computing system of  claim 1 , wherein the set of message types included in the context database are determined based on the classification model and an input obtained from a user via a graphical user interface. 
     
     
       19. The media of  claim 14 , wherein the data format of first expected context comprises an embedding format. 
     
     
       20. The method of  claim 9 , wherein the expected context representation is determined via machine learning.